Research On Task Scheduling Based On Load Balancing And Task Overtime Rate

In recent years, under the impetus of the academia and industry, cloud computing technology has been developed rapidly. The implementation of cloud computing technology in addition to rely on technology such as distributed computing and grid computing, but also depends on a key technology-load balancing technology. How to achieve load balancing resource access in order to improve the system’s overall processing capacity is one of the key issues of cloud computing. It is the most basic requirement for users of cloud computing services that tasks are not timed out. How to reduce tasks overtime rate by reasonable scheduling and resource management is also one of the important objectives of cloud computing research. With the expansion of the scale of cloud computing systems, the operating cost and energy consumption of cloud computing attract more and more attention. It is the development trend and direction of cloud computing technology to improve power efficiency and achieve green computing. Therefore, this thesis targets load balancing, tasks overtime rate and system energy consumption as the task scheduling optimization objectives. Research work done as follows:(1) This thesis introduces the characteristics and optimization goals of cloud computing task scheduling. This thesis also studies and compares the advantages and disadvantages of a wide range of classical task scheduling models. According to the above research results, we propose the task model, system model, load balancing model and energy consumption model. Then, this thesis briefly introduces several kinds of classic cloud computing task scheduling algorithms. These algorithms provide a theoretical basis for the analysis, comparison and improvement of algorithm.(2) This thesis mainly studies the centralized batch scheduling model. On the basis of this model, it still has a high rate of tasks overtime after speed optimization based on turnaround time. Then, this thesis comparatively studies an improved scheme which is based on deadline time. Under the heterogeneous environment, load balancing method based on the number of jobs can not accurately reflect the status of server load. Then, this thesis analyses a load balancing method based on the pre-execution time of servers, and this method improves the performance of the load balancing. Genetic algorithm is easy to fall into local solution. This thesis improves genetic algorithm by using the Metropolis criterion which is based on simulated annealing algorithm. It effectively improves the global search ability of genetic algorithm. And this thesis adopts dual fitness function to optimize load balancing and system energy consumption, so the population can evolve toward the direction which the performance of load balancing is good and system energy consumption is low.(3) The centralized batch scheduling model has only one scheduler for task scheduling. Large and complex cloud computing environment is easy to cause scheduling system failure. The running time of simulated annealing genetic algorithm is so long that it can not guarantee the real-time requirement of task execution. For the above shortcomings, this thesis comparatively studies a hierarchical real-time scheduling model and an algorithm named LTS which is based on this scheduling model. The simulation results show that LTS algorithm has certain advantage in reducing task overtime rate compared with simulated annealing genetic algorithm. At the same time, LTS algorithm maintains a good load balancing performance and effectively reduces the system energy consumption.